Robotics Software & Systems Test Engineer

• Reduced training time for transfer-learning vision classifiers (frozen DINOv2 ViT-B/14 and ResNet-50 backbones) from over 5 hours to under 4 minutes via fp16 autocast, cached backbone embeddings, and pipelined CPU preprocessing against GPU inference.
• Built perception for automated seal-quality inspection, detecting wrinkling by scoring six plate regions independently rather than the plate as a whole — raising good-seal recall from 4.52% to 100% and enabling automated quarantine of defective plates before downstream processing.
• Programmed HighRes Flexpod and Prime liquid handlers to run the full inspection loop unattended and synthesize per-well training data on demand, dispensing controlled concentration gradients and artifacts across 300,000+ evaluations.
• Characterized classifier robustness under varying illumination with lux-calibrated Philips Hue rigs, exposing a 9-lux shift that collapsed plate pass rate from 100% to 71.29%, and partnered with mechanical engineering to translate the findings into camera, lighting, and transparent-nest design.

Network Operations Technician

• Resolved technical support and debugging requests for Black Box's Taiwan office through ServiceNow, triaging and documenting incidents across a hybrid on-site and remote workflow.

Machine Learning Engineer

• Engineered an unsupervised learning model (K-Means clustering and PCA) in Python, scikit-learn, and Pandas to analyze high-dimensional telemetry from 5 million players across 40+ countries.
• Built a robust ETL pipeline to process raw accelerometer signals across x, y, z axes, applying noise reduction and normalization to classify motion primitives like dodging and jumping.
• Deployed the model into the production backend to generate real-time performance analytics, benchmarking user agility against professional player datasets for algorithmic skill feedback.

Tech Coordinator

• Architected a digital guide system scaling to 30,000+ annual users.
• Modernized on-premise infrastructure and SQL databases.

Technology Lead

• Led infrastructure for a hackathon of 100+ participants.
• Managed the website and submission workflows.

CNN Engineer

• Designed and trained a 3D convolutional neural network on volumetric DCE-MRI scans to predict pathological complete response (pCR) in triple-negative breast cancer patients.
• Built a data processing pipeline to load, normalize, and augment 3D MRI volumes from the BreastDCEDL dataset, integrating patient metadata for multi-modal input.
• Implemented HiResCAM interpretability overlays to generate faithful heatmaps of model attention, enabling clinicians to identify the tissue regions driving each pCR prediction.

Python · PyTorch · 3D CNN · DCE-MRI · HiResCAM

Quake

Accessibility

Real-time audio recognition app built on TensorFlow Lite, surfacing environmental sound as haptic feedback.

Audience Analyzer

Data Tooling

Python and Selenium tool for YouTube audience demographics; sold to a digital management firm.